Repository: SE-PINN
Summary: The mathematical properties of normality, symmetry, and orthogonality are integrated into a neural network in PyTorch via a custom loss function and a custom architectural layer so that its predictions respect these properties by design.
Figure: The prediction of the energy eigenvector (left) appears to converge, but the prediction of the energy eigenvalue (right) is slower. Visit the repository to see the same visualization when the model is constrained to respect symmetry in its predictions.
